💡 Building an AI we can trust, with Anthropic’s Dario Amodei

What it takes to build honest, safe AI

In a few years, we will rely on trustworthy AI systems as part of everyday life. That is the premise of my conversation with Anthropic’s co-founder and CEO, Dario Amodei.

Dario is a bona fide researcher: his papers have been cited 30,000 times in the past seven years. And as Anthropic’s Chief Executive, he is at the epicentre of an enormous explosion.

Dario is also very attentive to the problems of harm, and he will be one of the participants in the first major AI safety summit hosted by the U.K. government next week. I wanted to understand how Dario and Anthropic think about designing AI that is honest and harmless — and whether it is possible:

We have at least a few years of the current exponential curve — we may get to the point where AI systems can get broader and more creative than humans. […] Let’s assume we get right the rules of the road — [to build] safe, honest, harmless AI — I do think that everyone could have an AI assistant they really trust to help them make better decisions in their life.

Dario and I go into…

• The intricacies of creating LLMs and the insidious nature of imperfect systems,

• Anthropic’s approach to developing methods to peer into the black box of AI,

• One of the key tenants of safety: the AI being able to tell us what it does not know.

You can also listen to this conversation on Apple Podcasts, Spotify or anywhere else you get your podcasts.